Change Detection In Remote Sensing Images Based On FCM Fusion Of Multi-feature And Level Set Model

With the development of remote sensing technology,the deepening of research and more and more abundant satellite data,change detection application has been expanded gradually,but the methods of change detection are not perfect.The method of change detection can be classified into: supervised and unsupervised change detection.The former method is based on the actual ground data to obtain the training samples;the latter method directly performs the threshold segmentation of the difference data of different image without any additional information.The representation of supervised change detection is the multi-feature fusion change detection.The disadvantage is that most of them are based on equal weights and cannot reflect the primary and secondary characteristics of each feature.The unsupervised method has different threshold determination approaches,the results of various threshold methods also have advantages and disadvantages with the different images.In response to these problems,the specific research content of this article mainly as followed:(1)In unsupervised change detection,the level set theory is introduced,which can avoid the threshold selection of the algebra operation.Improved distance regularization level set model(DRLSE),Region-scalable Fitting level set model(RSF),and local clustering variational level set model(LCVLS)are applied to the different image,then the results are compared with the traditional adaptive threshold method.The results show that the accuracy of LCVLS model is higher than the RSF model and the RSF model is higher than the improved DRLSE model,the traditional adaptive threshold method has the lowest accuracy.(2)In supervised change detection,FCM(Fuzzy c-means)object-oriented change detection based on fusion of multi-feature is proposed.Firstly,the weights of the spectral features,Gabor features and GLCM features are calculated using the relief algorithm,then the primary and secondary importance of each feature was indicated Moreover,three experiments was performed with individual features,direct weighted fusion,and weighted FCM fusion respectively.Experiments show that the weighted FCM fusion has higher accuracy than direct weighted fusion,and direct weighted fusion is higher than single feature.(3)The combination of weighted FCM multi-feature fusion and the optimum LCVLS model are used to process the fusion features in this paper.Experiments show that the overall accuracy of the combined model is better than that the single one in unsupervised and supervised change detection.What's more,the combined model has the best performance in the reducing of Commission ratio and Omission ratio,which indicates that the combined model is conducive to improving the accuracy of change detection.